blob: c2e58880207faa268e32541cdb68f7d0e7f6f9b2 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/nommu.c
3 *
4 * Replacement code for mm functions to support CPU's that don't
5 * have any form of memory management unit (thus no virtual memory).
6 *
7 * See Documentation/nommu-mmap.txt
8 *
David Howells8feae132009-01-08 12:04:47 +00009 * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11 * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12 * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com>
Paul Mundt29c185e2010-12-24 12:08:30 +090013 * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014 */
15
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070016#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040018#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/mm.h>
Davidlohr Bueso615d6e82014-04-07 15:37:25 -070020#include <linux/vmacache.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/mman.h>
22#include <linux/swap.h>
23#include <linux/file.h>
24#include <linux/highmem.h>
25#include <linux/pagemap.h>
26#include <linux/slab.h>
27#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/blkdev.h>
29#include <linux/backing-dev.h>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -070030#include <linux/compiler.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/mount.h>
32#include <linux/personality.h>
33#include <linux/security.h>
34#include <linux/syscalls.h>
Al Viro120a7952010-10-30 02:54:44 -040035#include <linux/audit.h>
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070036#include <linux/printk.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38#include <asm/uaccess.h>
39#include <asm/tlb.h>
40#include <asm/tlbflush.h>
Bernd Schmidteb8cdec2009-09-21 17:03:57 -070041#include <asm/mmu_context.h>
David Howells8feae132009-01-08 12:04:47 +000042#include "internal.h"
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044void *high_memory;
Arnd Bergmann944b6872015-02-05 12:25:12 -080045EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070046struct page *mem_map;
47unsigned long max_mapnr;
gchen gchen5b8bf302015-03-12 16:26:05 -070048EXPORT_SYMBOL(max_mapnr);
Hugh Dickins4266c972009-09-23 17:05:53 +010049unsigned long highest_memmap_pfn;
David Howellsfc4d5c22009-05-06 16:03:05 -070050int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051int heap_stack_gap = 0;
52
David Howells33e5d7692009-04-02 16:56:32 -070053atomic_long_t mmap_pages_allocated;
David Howells8feae132009-01-08 12:04:47 +000054
Linus Torvalds1da177e2005-04-16 15:20:36 -070055EXPORT_SYMBOL(mem_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
David Howells8feae132009-01-08 12:04:47 +000057/* list of mapped, potentially shareable regions */
58static struct kmem_cache *vm_region_jar;
59struct rb_root nommu_region_tree = RB_ROOT;
60DECLARE_RWSEM(nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040062const struct vm_operations_struct generic_file_vm_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070063};
64
65/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 * Return the total memory allocated for this pointer, not
67 * just what the caller asked for.
68 *
69 * Doesn't have to be accurate, i.e. may have races.
70 */
71unsigned int kobjsize(const void *objp)
72{
73 struct page *page;
74
Michael Hennerich4016a132008-04-28 02:13:38 -070075 /*
76 * If the object we have should not have ksize performed on it,
77 * return size of 0
78 */
Paul Mundt5a1603b2008-06-12 16:29:55 +090079 if (!objp || !virt_addr_valid(objp))
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070080 return 0;
81
82 page = virt_to_head_page(objp);
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070083
84 /*
85 * If the allocator sets PageSlab, we know the pointer came from
86 * kmalloc().
87 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 if (PageSlab(page))
89 return ksize(objp);
90
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070091 /*
Paul Mundtab2e83e2009-01-08 12:04:48 +000092 * If it's not a compound page, see if we have a matching VMA
93 * region. This test is intentionally done in reverse order,
94 * so if there's no VMA, we still fall through and hand back
95 * PAGE_SIZE for 0-order pages.
96 */
97 if (!PageCompound(page)) {
98 struct vm_area_struct *vma;
99
100 vma = find_vma(current->mm, (unsigned long)objp);
101 if (vma)
102 return vma->vm_end - vma->vm_start;
103 }
104
105 /*
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700106 * The ksize() function is only guaranteed to work for pointers
Paul Mundt5a1603b2008-06-12 16:29:55 +0900107 * returned by kmalloc(). So handle arbitrary pointers here.
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700108 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900109 return PAGE_SIZE << compound_order(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110}
111
Michel Lespinasse28a35712013-02-22 16:35:55 -0800112long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
113 unsigned long start, unsigned long nr_pages,
114 unsigned int foll_flags, struct page **pages,
115 struct vm_area_struct **vmas, int *nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
Sonic Zhang910e46d2006-09-27 01:50:17 -0700117 struct vm_area_struct *vma;
David Howells7b4d5b82006-09-27 01:50:18 -0700118 unsigned long vm_flags;
119 int i;
120
121 /* calculate required read or write permissions.
Hugh Dickins58fa8792009-09-21 17:03:31 -0700122 * If FOLL_FORCE is set, we only require the "MAY" flags.
David Howells7b4d5b82006-09-27 01:50:18 -0700123 */
Hugh Dickins58fa8792009-09-21 17:03:31 -0700124 vm_flags = (foll_flags & FOLL_WRITE) ?
125 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
126 vm_flags &= (foll_flags & FOLL_FORCE) ?
127 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
Peter Zijlstra9d737772009-06-25 11:58:55 +0200129 for (i = 0; i < nr_pages; i++) {
David Howells7561e8c2010-03-25 16:48:38 +0000130 vma = find_vma(mm, start);
David Howells7b4d5b82006-09-27 01:50:18 -0700131 if (!vma)
132 goto finish_or_fault;
133
134 /* protect what we can, including chardevs */
Hugh Dickins1c3aff12009-09-21 17:03:24 -0700135 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
136 !(vm_flags & vma->vm_flags))
David Howells7b4d5b82006-09-27 01:50:18 -0700137 goto finish_or_fault;
Sonic Zhang910e46d2006-09-27 01:50:17 -0700138
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 if (pages) {
140 pages[i] = virt_to_page(start);
141 if (pages[i])
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300142 get_page(pages[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 }
144 if (vmas)
Sonic Zhang910e46d2006-09-27 01:50:17 -0700145 vmas[i] = vma;
David Howellse1ee65d2010-03-25 16:48:44 +0000146 start = (start + PAGE_SIZE) & PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 }
David Howells7b4d5b82006-09-27 01:50:18 -0700148
149 return i;
150
151finish_or_fault:
152 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153}
Nick Pigginb291f002008-10-18 20:26:44 -0700154
Nick Pigginb291f002008-10-18 20:26:44 -0700155/*
156 * get a list of pages in an address range belonging to the specified process
157 * and indicate the VMA that covers each page
158 * - this is potentially dodgy as we may end incrementing the page count of a
159 * slab page or a secondary page from a compound page
160 * - don't permit access to VMAs that don't support it, such as I/O mappings
161 */
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200162long get_user_pages(unsigned long start, unsigned long nr_pages,
Michel Lespinasse28a35712013-02-22 16:35:55 -0800163 int write, int force, struct page **pages,
164 struct vm_area_struct **vmas)
Nick Pigginb291f002008-10-18 20:26:44 -0700165{
166 int flags = 0;
167
168 if (write)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700169 flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700170 if (force)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700171 flags |= FOLL_FORCE;
Nick Pigginb291f002008-10-18 20:26:44 -0700172
Dave Hansencde70142016-02-12 13:01:55 -0800173 return __get_user_pages(current, current->mm, start, nr_pages, flags,
174 pages, vmas, NULL);
Nick Pigginb291f002008-10-18 20:26:44 -0700175}
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200176EXPORT_SYMBOL(get_user_pages);
Greg Ungerer66aa2b42005-09-12 11:18:10 +1000177
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200178long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
Dave Hansencde70142016-02-12 13:01:55 -0800179 int write, int force, struct page **pages,
180 int *locked)
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800181{
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200182 return get_user_pages(start, nr_pages, write, force, pages, NULL);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800183}
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200184EXPORT_SYMBOL(get_user_pages_locked);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800185
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800186long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
187 unsigned long start, unsigned long nr_pages,
188 int write, int force, struct page **pages,
189 unsigned int gup_flags)
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800190{
191 long ret;
192 down_read(&mm->mmap_sem);
Dave Hansencde70142016-02-12 13:01:55 -0800193 ret = __get_user_pages(tsk, mm, start, nr_pages, gup_flags, pages,
194 NULL, NULL);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800195 up_read(&mm->mmap_sem);
196 return ret;
197}
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800198EXPORT_SYMBOL(__get_user_pages_unlocked);
199
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200200long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800201 int write, int force, struct page **pages)
202{
Dave Hansencde70142016-02-12 13:01:55 -0800203 return __get_user_pages_unlocked(current, current->mm, start, nr_pages,
204 write, force, pages, 0);
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800205}
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200206EXPORT_SYMBOL(get_user_pages_unlocked);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800207
Paul Mundtdfc2f912009-06-26 04:31:57 +0900208/**
209 * follow_pfn - look up PFN at a user virtual address
210 * @vma: memory mapping
211 * @address: user virtual address
212 * @pfn: location to store found PFN
213 *
214 * Only IO mappings and raw PFN mappings are allowed.
215 *
216 * Returns zero and the pfn at @pfn on success, -ve otherwise.
217 */
218int follow_pfn(struct vm_area_struct *vma, unsigned long address,
219 unsigned long *pfn)
220{
221 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
222 return -EINVAL;
223
224 *pfn = address >> PAGE_SHIFT;
225 return 0;
226}
227EXPORT_SYMBOL(follow_pfn);
228
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700229LIST_HEAD(vmap_area_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800231void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
233 kfree(addr);
234}
Paul Mundtb5073172007-07-21 04:37:25 -0700235EXPORT_SYMBOL(vfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
Al Virodd0fc662005-10-07 07:46:04 +0100237void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
239 /*
Robert P. J. Day85186092007-10-19 23:11:38 +0200240 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
241 * returns only a logical address.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 */
Nick Piggin84097512006-03-22 00:08:34 -0800243 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244}
Paul Mundtb5073172007-07-21 04:37:25 -0700245EXPORT_SYMBOL(__vmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
Paul Mundtf905bc42008-02-04 22:29:59 -0800247void *vmalloc_user(unsigned long size)
248{
249 void *ret;
250
251 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
252 PAGE_KERNEL);
253 if (ret) {
254 struct vm_area_struct *vma;
255
256 down_write(&current->mm->mmap_sem);
257 vma = find_vma(current->mm, (unsigned long)ret);
258 if (vma)
259 vma->vm_flags |= VM_USERMAP;
260 up_write(&current->mm->mmap_sem);
261 }
262
263 return ret;
264}
265EXPORT_SYMBOL(vmalloc_user);
266
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800267struct page *vmalloc_to_page(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268{
269 return virt_to_page(addr);
270}
Paul Mundtb5073172007-07-21 04:37:25 -0700271EXPORT_SYMBOL(vmalloc_to_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800273unsigned long vmalloc_to_pfn(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274{
275 return page_to_pfn(virt_to_page(addr));
276}
Paul Mundtb5073172007-07-21 04:37:25 -0700277EXPORT_SYMBOL(vmalloc_to_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
279long vread(char *buf, char *addr, unsigned long count)
280{
Chen Gang9bde9162013-07-03 15:02:36 -0700281 /* Don't allow overflow */
282 if ((unsigned long) buf + count < count)
283 count = -(unsigned long) buf;
284
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 memcpy(buf, addr, count);
286 return count;
287}
288
289long vwrite(char *buf, char *addr, unsigned long count)
290{
291 /* Don't allow overflow */
292 if ((unsigned long) addr + count < count)
293 count = -(unsigned long) addr;
294
295 memcpy(addr, buf, count);
Choi Gi-yongac714902014-04-07 15:37:36 -0700296 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297}
298
299/*
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900300 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 *
302 * @size: allocation size
303 *
304 * Allocate enough pages to cover @size from the page level
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900305 * allocator and map them into contiguous kernel virtual space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200307 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 * use __vmalloc() instead.
309 */
310void *vmalloc(unsigned long size)
311{
312 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
313}
Andrew Mortonf6138882006-02-28 16:59:18 -0800314EXPORT_SYMBOL(vmalloc);
315
Dave Younge1ca7782010-10-26 14:22:06 -0700316/*
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900317 * vzalloc - allocate virtually contiguous memory with zero fill
Dave Younge1ca7782010-10-26 14:22:06 -0700318 *
319 * @size: allocation size
320 *
321 * Allocate enough pages to cover @size from the page level
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900322 * allocator and map them into contiguous kernel virtual space.
Dave Younge1ca7782010-10-26 14:22:06 -0700323 * The memory allocated is set to zero.
324 *
325 * For tight control over page level allocator and protection flags
326 * use __vmalloc() instead.
327 */
328void *vzalloc(unsigned long size)
329{
330 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
331 PAGE_KERNEL);
332}
333EXPORT_SYMBOL(vzalloc);
334
335/**
336 * vmalloc_node - allocate memory on a specific node
337 * @size: allocation size
338 * @node: numa node
339 *
340 * Allocate enough pages to cover @size from the page level
341 * allocator and map them into contiguous kernel virtual space.
342 *
343 * For tight control over page level allocator and protection flags
344 * use __vmalloc() instead.
345 */
Andrew Mortonf6138882006-02-28 16:59:18 -0800346void *vmalloc_node(unsigned long size, int node)
347{
348 return vmalloc(size);
349}
Paul Mundt9a14f652010-12-24 11:50:34 +0900350EXPORT_SYMBOL(vmalloc_node);
Dave Younge1ca7782010-10-26 14:22:06 -0700351
352/**
353 * vzalloc_node - allocate memory on a specific node with zero fill
354 * @size: allocation size
355 * @node: numa node
356 *
357 * Allocate enough pages to cover @size from the page level
358 * allocator and map them into contiguous kernel virtual space.
359 * The memory allocated is set to zero.
360 *
361 * For tight control over page level allocator and protection flags
362 * use __vmalloc() instead.
363 */
364void *vzalloc_node(unsigned long size, int node)
365{
366 return vzalloc(size);
367}
368EXPORT_SYMBOL(vzalloc_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369
Paul Mundt1af446e2008-08-04 16:01:47 +0900370#ifndef PAGE_KERNEL_EXEC
371# define PAGE_KERNEL_EXEC PAGE_KERNEL
372#endif
373
374/**
375 * vmalloc_exec - allocate virtually contiguous, executable memory
376 * @size: allocation size
377 *
378 * Kernel-internal function to allocate enough pages to cover @size
379 * the page level allocator and map them into contiguous and
380 * executable kernel virtual space.
381 *
382 * For tight control over page level allocator and protection flags
383 * use __vmalloc() instead.
384 */
385
386void *vmalloc_exec(unsigned long size)
387{
388 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
389}
390
Paul Mundtb5073172007-07-21 04:37:25 -0700391/**
392 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 * @size: allocation size
394 *
395 * Allocate enough 32bit PA addressable pages to cover @size from the
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900396 * page level allocator and map them into contiguous kernel virtual space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 */
398void *vmalloc_32(unsigned long size)
399{
400 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
401}
Paul Mundtb5073172007-07-21 04:37:25 -0700402EXPORT_SYMBOL(vmalloc_32);
403
404/**
405 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
406 * @size: allocation size
407 *
408 * The resulting memory area is 32bit addressable and zeroed so it can be
409 * mapped to userspace without leaking data.
Paul Mundtf905bc42008-02-04 22:29:59 -0800410 *
411 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
412 * remap_vmalloc_range() are permissible.
Paul Mundtb5073172007-07-21 04:37:25 -0700413 */
414void *vmalloc_32_user(unsigned long size)
415{
Paul Mundtf905bc42008-02-04 22:29:59 -0800416 /*
417 * We'll have to sort out the ZONE_DMA bits for 64-bit,
418 * but for now this can simply use vmalloc_user() directly.
419 */
420 return vmalloc_user(size);
Paul Mundtb5073172007-07-21 04:37:25 -0700421}
422EXPORT_SYMBOL(vmalloc_32_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423
424void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
425{
426 BUG();
427 return NULL;
428}
Paul Mundtb5073172007-07-21 04:37:25 -0700429EXPORT_SYMBOL(vmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800431void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
433 BUG();
434}
Paul Mundtb5073172007-07-21 04:37:25 -0700435EXPORT_SYMBOL(vunmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
Paul Mundteb6434d2009-01-21 17:45:47 +0900437void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
438{
439 BUG();
440 return NULL;
441}
442EXPORT_SYMBOL(vm_map_ram);
443
444void vm_unmap_ram(const void *mem, unsigned int count)
445{
446 BUG();
447}
448EXPORT_SYMBOL(vm_unmap_ram);
449
450void vm_unmap_aliases(void)
451{
452}
453EXPORT_SYMBOL_GPL(vm_unmap_aliases);
454
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455/*
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700456 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
457 * have one.
458 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -0700459void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700460{
461}
462
Paul Mundt29c185e2010-12-24 12:08:30 +0900463/**
464 * alloc_vm_area - allocate a range of kernel address space
465 * @size: size of the area
466 *
467 * Returns: NULL on failure, vm_struct on success
468 *
469 * This function reserves a range of kernel address space, and
470 * allocates pagetables to map that range. No actual mappings
471 * are created. If the kernel address space is not shared
472 * between processes, it syncs the pagetable across all
473 * processes.
474 */
David Vrabelcd129092011-09-29 16:53:32 +0100475struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Paul Mundt29c185e2010-12-24 12:08:30 +0900476{
477 BUG();
478 return NULL;
479}
480EXPORT_SYMBOL_GPL(alloc_vm_area);
481
482void free_vm_area(struct vm_struct *area)
483{
484 BUG();
485}
486EXPORT_SYMBOL_GPL(free_vm_area);
487
Paul Mundtb5073172007-07-21 04:37:25 -0700488int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
489 struct page *page)
490{
491 return -EINVAL;
492}
493EXPORT_SYMBOL(vm_insert_page);
494
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700495/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 * sys_brk() for the most part doesn't need the global kernel
497 * lock, except when an application is doing something nasty
498 * like trying to un-brk an area that has already been mapped
499 * to a regular file. in this case, the unmapping will need
500 * to invoke file system routines that need the global lock.
501 */
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100502SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503{
504 struct mm_struct *mm = current->mm;
505
506 if (brk < mm->start_brk || brk > mm->context.end_brk)
507 return mm->brk;
508
509 if (mm->brk == brk)
510 return mm->brk;
511
512 /*
513 * Always allow shrinking brk
514 */
515 if (brk <= mm->brk) {
516 mm->brk = brk;
517 return brk;
518 }
519
520 /*
521 * Ok, looks good - let it rip.
522 */
Mike Frysingercfe79c02010-01-06 17:23:23 +0000523 flush_icache_range(mm->brk, brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 return mm->brk = brk;
525}
526
David Howells8feae132009-01-08 12:04:47 +0000527/*
528 * initialise the VMA and region record slabs
529 */
530void __init mmap_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700532 int ret;
533
Tejun Heo908c7f12014-09-08 09:51:29 +0900534 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700535 VM_BUG_ON(ret);
Vladimir Davydov5d097052016-01-14 15:18:21 -0800536 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
David Howells8feae132009-01-08 12:04:47 +0000537}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538
David Howells8feae132009-01-08 12:04:47 +0000539/*
540 * validate the region tree
541 * - the caller must hold the region lock
542 */
543#ifdef CONFIG_DEBUG_NOMMU_REGIONS
544static noinline void validate_nommu_regions(void)
545{
546 struct vm_region *region, *last;
547 struct rb_node *p, *lastp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
David Howells8feae132009-01-08 12:04:47 +0000549 lastp = rb_first(&nommu_region_tree);
550 if (!lastp)
551 return;
552
553 last = rb_entry(lastp, struct vm_region, vm_rb);
Geliang Tangc9427bc2015-11-05 18:48:38 -0800554 BUG_ON(last->vm_end <= last->vm_start);
555 BUG_ON(last->vm_top < last->vm_end);
David Howells8feae132009-01-08 12:04:47 +0000556
557 while ((p = rb_next(lastp))) {
558 region = rb_entry(p, struct vm_region, vm_rb);
559 last = rb_entry(lastp, struct vm_region, vm_rb);
560
Geliang Tangc9427bc2015-11-05 18:48:38 -0800561 BUG_ON(region->vm_end <= region->vm_start);
562 BUG_ON(region->vm_top < region->vm_end);
563 BUG_ON(region->vm_start < last->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000564
565 lastp = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 }
567}
David Howells8feae132009-01-08 12:04:47 +0000568#else
David Howells33e5d7692009-04-02 16:56:32 -0700569static void validate_nommu_regions(void)
570{
571}
David Howells8feae132009-01-08 12:04:47 +0000572#endif
573
574/*
575 * add a region into the global tree
576 */
577static void add_nommu_region(struct vm_region *region)
578{
579 struct vm_region *pregion;
580 struct rb_node **p, *parent;
581
582 validate_nommu_regions();
583
David Howells8feae132009-01-08 12:04:47 +0000584 parent = NULL;
585 p = &nommu_region_tree.rb_node;
586 while (*p) {
587 parent = *p;
588 pregion = rb_entry(parent, struct vm_region, vm_rb);
589 if (region->vm_start < pregion->vm_start)
590 p = &(*p)->rb_left;
591 else if (region->vm_start > pregion->vm_start)
592 p = &(*p)->rb_right;
593 else if (pregion == region)
594 return;
595 else
596 BUG();
597 }
598
599 rb_link_node(&region->vm_rb, parent, p);
600 rb_insert_color(&region->vm_rb, &nommu_region_tree);
601
602 validate_nommu_regions();
603}
604
605/*
606 * delete a region from the global tree
607 */
608static void delete_nommu_region(struct vm_region *region)
609{
610 BUG_ON(!nommu_region_tree.rb_node);
611
612 validate_nommu_regions();
613 rb_erase(&region->vm_rb, &nommu_region_tree);
614 validate_nommu_regions();
615}
616
617/*
618 * free a contiguous series of pages
619 */
620static void free_page_series(unsigned long from, unsigned long to)
621{
622 for (; from < to; from += PAGE_SIZE) {
623 struct page *page = virt_to_page(from);
624
David Howells33e5d7692009-04-02 16:56:32 -0700625 atomic_long_dec(&mmap_pages_allocated);
David Howells8feae132009-01-08 12:04:47 +0000626 put_page(page);
627 }
628}
629
630/*
631 * release a reference to a region
David Howells33e5d7692009-04-02 16:56:32 -0700632 * - the caller must hold the region semaphore for writing, which this releases
Paul Mundtdd8632a2009-01-08 12:04:47 +0000633 * - the region may not have been added to the tree yet, in which case vm_top
David Howells8feae132009-01-08 12:04:47 +0000634 * will equal vm_start
635 */
636static void __put_nommu_region(struct vm_region *region)
637 __releases(nommu_region_sem)
638{
David Howells8feae132009-01-08 12:04:47 +0000639 BUG_ON(!nommu_region_tree.rb_node);
640
David Howells1e2ae592010-01-15 17:01:33 -0800641 if (--region->vm_usage == 0) {
Paul Mundtdd8632a2009-01-08 12:04:47 +0000642 if (region->vm_top > region->vm_start)
David Howells8feae132009-01-08 12:04:47 +0000643 delete_nommu_region(region);
644 up_write(&nommu_region_sem);
645
646 if (region->vm_file)
647 fput(region->vm_file);
648
649 /* IO memory and memory shared directly out of the pagecache
650 * from ramfs/tmpfs mustn't be released here */
Leon Romanovsky22cc8772015-06-24 16:57:47 -0700651 if (region->vm_flags & VM_MAPPED_COPY)
Paul Mundtdd8632a2009-01-08 12:04:47 +0000652 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000653 kmem_cache_free(vm_region_jar, region);
654 } else {
655 up_write(&nommu_region_sem);
656 }
657}
658
659/*
660 * release a reference to a region
661 */
662static void put_nommu_region(struct vm_region *region)
663{
664 down_write(&nommu_region_sem);
665 __put_nommu_region(region);
666}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
David Howells30340972006-09-27 01:50:20 -0700668/*
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700669 * update protection on a vma
670 */
671static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
672{
673#ifdef CONFIG_MPU
674 struct mm_struct *mm = vma->vm_mm;
675 long start = vma->vm_start & PAGE_MASK;
676 while (start < vma->vm_end) {
677 protect_page(mm, start, flags);
678 start += PAGE_SIZE;
679 }
680 update_protections(mm);
681#endif
682}
683
684/*
David Howells30340972006-09-27 01:50:20 -0700685 * add a VMA into a process's mm_struct in the appropriate place in the list
David Howells8feae132009-01-08 12:04:47 +0000686 * and tree and add to the address space's page tree also if not an anonymous
687 * page
David Howells30340972006-09-27 01:50:20 -0700688 * - should be called with mm->mmap_sem held writelocked
689 */
David Howells8feae132009-01-08 12:04:47 +0000690static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
David Howells30340972006-09-27 01:50:20 -0700691{
Namhyung Kim6038def2011-05-24 17:11:22 -0700692 struct vm_area_struct *pvma, *prev;
David Howells8feae132009-01-08 12:04:47 +0000693 struct address_space *mapping;
Namhyung Kim6038def2011-05-24 17:11:22 -0700694 struct rb_node **p, *parent, *rb_prev;
David Howells30340972006-09-27 01:50:20 -0700695
David Howells8feae132009-01-08 12:04:47 +0000696 BUG_ON(!vma->vm_region);
697
698 mm->map_count++;
699 vma->vm_mm = mm;
700
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700701 protect_vma(vma, vma->vm_flags);
702
David Howells8feae132009-01-08 12:04:47 +0000703 /* add the VMA to the mapping */
704 if (vma->vm_file) {
705 mapping = vma->vm_file->f_mapping;
706
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800707 i_mmap_lock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000708 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700709 vma_interval_tree_insert(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000710 flush_dcache_mmap_unlock(mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800711 i_mmap_unlock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000712 }
713
714 /* add the VMA to the tree */
Namhyung Kim6038def2011-05-24 17:11:22 -0700715 parent = rb_prev = NULL;
David Howells8feae132009-01-08 12:04:47 +0000716 p = &mm->mm_rb.rb_node;
717 while (*p) {
718 parent = *p;
719 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
720
721 /* sort by: start addr, end addr, VMA struct addr in that order
722 * (the latter is necessary as we may get identical VMAs) */
723 if (vma->vm_start < pvma->vm_start)
724 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700725 else if (vma->vm_start > pvma->vm_start) {
726 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000727 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700728 } else if (vma->vm_end < pvma->vm_end)
David Howells8feae132009-01-08 12:04:47 +0000729 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700730 else if (vma->vm_end > pvma->vm_end) {
731 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000732 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700733 } else if (vma < pvma)
David Howells8feae132009-01-08 12:04:47 +0000734 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700735 else if (vma > pvma) {
736 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000737 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700738 } else
David Howells8feae132009-01-08 12:04:47 +0000739 BUG();
740 }
741
742 rb_link_node(&vma->vm_rb, parent, p);
743 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
744
745 /* add VMA to the VMA list also */
Namhyung Kim6038def2011-05-24 17:11:22 -0700746 prev = NULL;
747 if (rb_prev)
748 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
David Howells30340972006-09-27 01:50:20 -0700749
Namhyung Kim6038def2011-05-24 17:11:22 -0700750 __vma_link_list(mm, vma, prev, parent);
David Howells8feae132009-01-08 12:04:47 +0000751}
752
753/*
754 * delete a VMA from its owning mm_struct and address space
755 */
756static void delete_vma_from_mm(struct vm_area_struct *vma)
757{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700758 int i;
David Howells8feae132009-01-08 12:04:47 +0000759 struct address_space *mapping;
760 struct mm_struct *mm = vma->vm_mm;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700761 struct task_struct *curr = current;
David Howells8feae132009-01-08 12:04:47 +0000762
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700763 protect_vma(vma, 0);
764
David Howells8feae132009-01-08 12:04:47 +0000765 mm->map_count--;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700766 for (i = 0; i < VMACACHE_SIZE; i++) {
767 /* if the vma is cached, invalidate the entire cache */
768 if (curr->vmacache[i] == vma) {
Steven Miaoe020d5b2014-06-23 13:22:02 -0700769 vmacache_invalidate(mm);
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700770 break;
771 }
772 }
David Howells8feae132009-01-08 12:04:47 +0000773
774 /* remove the VMA from the mapping */
775 if (vma->vm_file) {
776 mapping = vma->vm_file->f_mapping;
777
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800778 i_mmap_lock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000779 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700780 vma_interval_tree_remove(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000781 flush_dcache_mmap_unlock(mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800782 i_mmap_unlock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000783 }
784
785 /* remove from the MM's tree and list */
786 rb_erase(&vma->vm_rb, &mm->mm_rb);
Namhyung Kimb951bf22011-05-24 17:11:23 -0700787
788 if (vma->vm_prev)
789 vma->vm_prev->vm_next = vma->vm_next;
790 else
791 mm->mmap = vma->vm_next;
792
793 if (vma->vm_next)
794 vma->vm_next->vm_prev = vma->vm_prev;
David Howells8feae132009-01-08 12:04:47 +0000795}
796
797/*
798 * destroy a VMA record
799 */
800static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
801{
David Howells8feae132009-01-08 12:04:47 +0000802 if (vma->vm_ops && vma->vm_ops->close)
803 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700804 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +0000805 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +0000806 put_nommu_region(vma->vm_region);
807 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700808}
809
810/*
811 * look up the first VMA in which addr resides, NULL if none
812 * - should be called with mm->mmap_sem at least held readlocked
813 */
814struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
815{
David Howells8feae132009-01-08 12:04:47 +0000816 struct vm_area_struct *vma;
David Howells30340972006-09-27 01:50:20 -0700817
David Howells8feae132009-01-08 12:04:47 +0000818 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700819 vma = vmacache_find(mm, addr);
820 if (likely(vma))
David Howells8feae132009-01-08 12:04:47 +0000821 return vma;
822
Namhyung Kime922c4c2011-05-24 17:11:24 -0700823 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000824 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700825 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000826 if (vma->vm_start > addr)
827 return NULL;
828 if (vma->vm_end > addr) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700829 vmacache_update(addr, vma);
David Howells8feae132009-01-08 12:04:47 +0000830 return vma;
831 }
David Howells30340972006-09-27 01:50:20 -0700832 }
833
David Howells30340972006-09-27 01:50:20 -0700834 return NULL;
835}
836EXPORT_SYMBOL(find_vma);
837
838/*
David Howells930e6522006-09-27 01:50:22 -0700839 * find a VMA
840 * - we don't extend stack VMAs under NOMMU conditions
841 */
842struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
843{
David Howells7561e8c2010-03-25 16:48:38 +0000844 return find_vma(mm, addr);
David Howells930e6522006-09-27 01:50:22 -0700845}
846
David Howells8feae132009-01-08 12:04:47 +0000847/*
848 * expand a stack to a given address
849 * - not supported under NOMMU conditions
850 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700851int expand_stack(struct vm_area_struct *vma, unsigned long address)
852{
853 return -ENOMEM;
854}
855
David Howells930e6522006-09-27 01:50:22 -0700856/*
David Howells6fa5f802006-09-27 01:50:21 -0700857 * look up the first VMA exactly that exactly matches addr
858 * - should be called with mm->mmap_sem at least held readlocked
859 */
David Howells8feae132009-01-08 12:04:47 +0000860static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
861 unsigned long addr,
862 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863{
864 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000865 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866
David Howells8feae132009-01-08 12:04:47 +0000867 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700868 vma = vmacache_find_exact(mm, addr, end);
869 if (vma)
David Howells8feae132009-01-08 12:04:47 +0000870 return vma;
871
Namhyung Kime922c4c2011-05-24 17:11:24 -0700872 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000873 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700874 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000875 if (vma->vm_start < addr)
876 continue;
877 if (vma->vm_start > addr)
878 return NULL;
879 if (vma->vm_end == end) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700880 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 return vma;
David Howells8feae132009-01-08 12:04:47 +0000882 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 }
884
885 return NULL;
886}
887
David Howells30340972006-09-27 01:50:20 -0700888/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 * determine whether a mapping should be permitted and, if so, what sort of
890 * mapping we're capable of supporting
891 */
892static int validate_mmap_request(struct file *file,
893 unsigned long addr,
894 unsigned long len,
895 unsigned long prot,
896 unsigned long flags,
897 unsigned long pgoff,
898 unsigned long *_capabilities)
899{
David Howells8feae132009-01-08 12:04:47 +0000900 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 int ret;
902
903 /* do the simple checks first */
Leon Romanovsky22cc8772015-06-24 16:57:47 -0700904 if (flags & MAP_FIXED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
907 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
908 (flags & MAP_TYPE) != MAP_SHARED)
909 return -EINVAL;
910
Mike Frysingerf81cff02006-12-06 12:02:59 +1000911 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 return -EINVAL;
913
Mike Frysingerf81cff02006-12-06 12:02:59 +1000914 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000915 rlen = PAGE_ALIGN(len);
916 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000917 return -ENOMEM;
918
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000920 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000921 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
923 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 /* files must support mmap */
Al Viro72c2d532013-09-22 16:27:52 -0400925 if (!file->f_op->mmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 return -ENODEV;
927
928 /* work out if what we've got could possibly be shared
929 * - we support chardevs that provide their own "memory"
930 * - we support files/blockdevs that are memory backed
931 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100932 if (file->f_op->mmap_capabilities) {
933 capabilities = file->f_op->mmap_capabilities(file);
934 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 /* no explicit capabilities set, so assume some
936 * defaults */
Al Viro496ad9a2013-01-23 17:07:38 -0500937 switch (file_inode(file)->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 case S_IFREG:
939 case S_IFBLK:
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100940 capabilities = NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 break;
942
943 case S_IFCHR:
944 capabilities =
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100945 NOMMU_MAP_DIRECT |
946 NOMMU_MAP_READ |
947 NOMMU_MAP_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 break;
949
950 default:
951 return -EINVAL;
952 }
953 }
954
955 /* eliminate any capabilities that we can't support on this
956 * device */
957 if (!file->f_op->get_unmapped_area)
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100958 capabilities &= ~NOMMU_MAP_DIRECT;
Al Viro6e242a12015-03-31 12:35:13 -0400959 if (!(file->f_mode & FMODE_CAN_READ))
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100960 capabilities &= ~NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
Graff Yang28d7a6a2009-08-18 14:11:17 -0700962 /* The file shall have been opened with read permission. */
963 if (!(file->f_mode & FMODE_READ))
964 return -EACCES;
965
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 if (flags & MAP_SHARED) {
967 /* do checks for writing, appending and locking */
968 if ((prot & PROT_WRITE) &&
969 !(file->f_mode & FMODE_WRITE))
970 return -EACCES;
971
Al Viro496ad9a2013-01-23 17:07:38 -0500972 if (IS_APPEND(file_inode(file)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 (file->f_mode & FMODE_WRITE))
974 return -EACCES;
975
Jeff Laytond7a06982014-03-10 09:54:15 -0400976 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 return -EAGAIN;
978
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100979 if (!(capabilities & NOMMU_MAP_DIRECT))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 return -ENODEV;
981
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 /* we mustn't privatise shared mappings */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100983 capabilities &= ~NOMMU_MAP_COPY;
Choi Gi-yongac714902014-04-07 15:37:36 -0700984 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 /* we're going to read the file into private memory we
986 * allocate */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100987 if (!(capabilities & NOMMU_MAP_COPY))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 return -ENODEV;
989
990 /* we don't permit a private writable mapping to be
991 * shared with the backing device */
992 if (prot & PROT_WRITE)
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100993 capabilities &= ~NOMMU_MAP_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 }
995
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100996 if (capabilities & NOMMU_MAP_DIRECT) {
997 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
998 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
999 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001000 ) {
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001001 capabilities &= ~NOMMU_MAP_DIRECT;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001002 if (flags & MAP_SHARED) {
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001003 pr_warn("MAP_SHARED not completely supported on !MMU\n");
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001004 return -EINVAL;
1005 }
1006 }
1007 }
1008
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 /* handle executable mappings and implied executable
1010 * mappings */
Eric W. Biederman90f85722015-06-29 14:42:03 -05001011 if (path_noexec(&file->f_path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 if (prot & PROT_EXEC)
1013 return -EPERM;
Choi Gi-yongac714902014-04-07 15:37:36 -07001014 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 /* handle implication of PROT_EXEC by PROT_READ */
1016 if (current->personality & READ_IMPLIES_EXEC) {
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001017 if (capabilities & NOMMU_MAP_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 prot |= PROT_EXEC;
1019 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001020 } else if ((prot & PROT_READ) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 (prot & PROT_EXEC) &&
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001022 !(capabilities & NOMMU_MAP_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 ) {
1024 /* backing file is not executable, try to copy */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001025 capabilities &= ~NOMMU_MAP_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001027 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 /* anonymous mappings are always memory backed and can be
1029 * privately mapped
1030 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001031 capabilities = NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
1033 /* handle PROT_EXEC implication by PROT_READ */
1034 if ((prot & PROT_READ) &&
1035 (current->personality & READ_IMPLIES_EXEC))
1036 prot |= PROT_EXEC;
1037 }
1038
1039 /* allow the security API to have its say */
Al Viroe5467852012-05-30 13:30:51 -04001040 ret = security_mmap_addr(addr);
1041 if (ret < 0)
1042 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
1044 /* looks okay */
1045 *_capabilities = capabilities;
1046 return 0;
1047}
1048
1049/*
1050 * we've determined that we can make the mapping, now translate what we
1051 * now know into VMA flags
1052 */
1053static unsigned long determine_vm_flags(struct file *file,
1054 unsigned long prot,
1055 unsigned long flags,
1056 unsigned long capabilities)
1057{
1058 unsigned long vm_flags;
1059
Dave Hansene6bfb702016-02-12 13:02:31 -08001060 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 /* vm_flags |= mm->def_flags; */
1062
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001063 if (!(capabilities & NOMMU_MAP_DIRECT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 /* attempt to share read-only copies of mapped file chunks */
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001065 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 if (file && !(prot & PROT_WRITE))
1067 vm_flags |= VM_MAYSHARE;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001068 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 /* overlay a shareable mapping on the backing device or inode
1070 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1071 * romfs/cramfs */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001072 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 if (flags & MAP_SHARED)
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001074 vm_flags |= VM_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 }
1076
1077 /* refuse to let anyone share private mappings with this process if
1078 * it's being traced - otherwise breakpoints set in it may interfere
1079 * with another untraced process
1080 */
Tejun Heoa288eec2011-06-17 16:50:37 +02001081 if ((flags & MAP_PRIVATE) && current->ptrace)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 vm_flags &= ~VM_MAYSHARE;
1083
1084 return vm_flags;
1085}
1086
1087/*
David Howells8feae132009-01-08 12:04:47 +00001088 * set up a shared mapping on a file (the driver or filesystem provides and
1089 * pins the storage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 */
David Howells8feae132009-01-08 12:04:47 +00001091static int do_mmap_shared_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092{
1093 int ret;
1094
1095 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001096 if (ret == 0) {
1097 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001098 return 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001099 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 if (ret != -ENOSYS)
1101 return ret;
1102
David Howells3fa30462010-03-23 13:35:21 -07001103 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1104 * opposed to tried but failed) so we can only give a suitable error as
1105 * it's not possible to make a private copy if MAP_SHARED was given */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 return -ENODEV;
1107}
1108
1109/*
1110 * set up a private mapping or an anonymous shared mapping
1111 */
David Howells8feae132009-01-08 12:04:47 +00001112static int do_mmap_private(struct vm_area_struct *vma,
1113 struct vm_region *region,
David Howells645d83c2009-09-24 15:13:10 +01001114 unsigned long len,
1115 unsigned long capabilities)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116{
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001117 unsigned long total, point;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 void *base;
David Howells8feae132009-01-08 12:04:47 +00001119 int ret, order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
1121 /* invoke the file's mapping function so that it can keep track of
1122 * shared mappings on devices or memory
1123 * - VM_MAYSHARE will be set if it may attempt to share
1124 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001125 if (capabilities & NOMMU_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001127 if (ret == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 /* shouldn't return success if we're not sharing */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001129 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1130 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001131 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 }
Paul Mundtdd8632a2009-01-08 12:04:47 +00001133 if (ret != -ENOSYS)
1134 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
1136 /* getting an ENOSYS error indicates that direct mmap isn't
1137 * possible (as opposed to tried but failed) so we'll try to
1138 * make a private copy of the data and map that instead */
1139 }
1140
David Howells8feae132009-01-08 12:04:47 +00001141
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 /* allocate some memory to hold the mapping
1143 * - note that this may not return a page-aligned address if the object
1144 * we're allocating is smaller than a page
1145 */
Bob Liuf67d9b12011-05-24 17:12:56 -07001146 order = get_order(len);
David Howells8feae132009-01-08 12:04:47 +00001147 total = 1 << order;
Bob Liuf67d9b12011-05-24 17:12:56 -07001148 point = len >> PAGE_SHIFT;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001149
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001150 /* we don't want to allocate a power-of-2 sized page set */
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001151 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001152 total = point;
David Howells8feae132009-01-08 12:04:47 +00001153
Joonsoo Kimda616532015-02-27 15:51:43 -08001154 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001155 if (!base)
1156 goto enomem;
David Howells8feae132009-01-08 12:04:47 +00001157
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001158 atomic_long_add(total, &mmap_pages_allocated);
1159
David Howells8feae132009-01-08 12:04:47 +00001160 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1161 region->vm_start = (unsigned long) base;
Bob Liuf67d9b12011-05-24 17:12:56 -07001162 region->vm_end = region->vm_start + len;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001163 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
David Howells8feae132009-01-08 12:04:47 +00001164
1165 vma->vm_start = region->vm_start;
1166 vma->vm_end = region->vm_start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167
1168 if (vma->vm_file) {
1169 /* read the contents of a file into the copy */
1170 mm_segment_t old_fs;
1171 loff_t fpos;
1172
1173 fpos = vma->vm_pgoff;
1174 fpos <<= PAGE_SHIFT;
1175
1176 old_fs = get_fs();
1177 set_fs(KERNEL_DS);
Al Viro6e242a12015-03-31 12:35:13 -04001178 ret = __vfs_read(vma->vm_file, base, len, &fpos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 set_fs(old_fs);
1180
1181 if (ret < 0)
1182 goto error_free;
1183
1184 /* clear the last little bit */
Bob Liuf67d9b12011-05-24 17:12:56 -07001185 if (ret < len)
1186 memset(base + ret, 0, len - ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 }
1189
1190 return 0;
1191
1192error_free:
Namhyung Kim7223bb42011-05-24 17:11:26 -07001193 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +00001194 region->vm_start = vma->vm_start = 0;
1195 region->vm_end = vma->vm_end = 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001196 region->vm_top = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 return ret;
1198
1199enomem:
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07001200 pr_err("Allocation of length %lu from process %d (%s) failed\n",
Greg Ungerer05ae6fa2009-01-13 17:30:22 +10001201 len, current->pid, current->comm);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001202 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 return -ENOMEM;
1204}
1205
1206/*
1207 * handle mapping creation for uClinux
1208 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001209unsigned long do_mmap(struct file *file,
1210 unsigned long addr,
1211 unsigned long len,
1212 unsigned long prot,
1213 unsigned long flags,
1214 vm_flags_t vm_flags,
1215 unsigned long pgoff,
1216 unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217{
David Howells8feae132009-01-08 12:04:47 +00001218 struct vm_area_struct *vma;
1219 struct vm_region *region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 struct rb_node *rb;
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001221 unsigned long capabilities, result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 int ret;
1223
Michel Lespinasse41badc12013-02-22 16:32:47 -08001224 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001225
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 /* decide whether we should attempt the mapping, and if so what sort of
1227 * mapping */
1228 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1229 &capabilities);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001230 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 return ret;
1232
David Howells06aab5a2009-09-24 12:33:48 +01001233 /* we ignore the address hint */
1234 addr = 0;
Bob Liuf67d9b12011-05-24 17:12:56 -07001235 len = PAGE_ALIGN(len);
David Howells06aab5a2009-09-24 12:33:48 +01001236
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 /* we've determined that we can make the mapping, now translate what we
1238 * now know into VMA flags */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001239 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
David Howells8feae132009-01-08 12:04:47 +00001241 /* we're going to need to record the mapping */
1242 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1243 if (!region)
1244 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245
David Howells8feae132009-01-08 12:04:47 +00001246 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1247 if (!vma)
1248 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
David Howells1e2ae592010-01-15 17:01:33 -08001250 region->vm_usage = 1;
David Howells8feae132009-01-08 12:04:47 +00001251 region->vm_flags = vm_flags;
1252 region->vm_pgoff = pgoff;
1253
Rik van Riel5beb4932010-03-05 13:42:07 -08001254 INIT_LIST_HEAD(&vma->anon_vma_chain);
David Howells8feae132009-01-08 12:04:47 +00001255 vma->vm_flags = vm_flags;
1256 vma->vm_pgoff = pgoff;
1257
1258 if (file) {
Al Virocb0942b2012-08-27 14:48:26 -04001259 region->vm_file = get_file(file);
1260 vma->vm_file = get_file(file);
David Howells8feae132009-01-08 12:04:47 +00001261 }
1262
1263 down_write(&nommu_region_sem);
1264
1265 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001267 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 * - shared mappings on character devices and memory backed files are
1269 * permitted to overlap inexactly as far as we are concerned for in
1270 * these cases, sharing is handled in the driver or filesystem rather
1271 * than here
1272 */
1273 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001274 struct vm_region *pregion;
1275 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
David Howells8feae132009-01-08 12:04:47 +00001277 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1278 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001279
David Howells8feae132009-01-08 12:04:47 +00001280 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1281 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282
David Howells8feae132009-01-08 12:04:47 +00001283 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 continue;
1285
1286 /* search for overlapping mappings on the same file */
Al Viro496ad9a2013-01-23 17:07:38 -05001287 if (file_inode(pregion->vm_file) !=
1288 file_inode(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 continue;
1290
David Howells8feae132009-01-08 12:04:47 +00001291 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 continue;
1293
David Howells8feae132009-01-08 12:04:47 +00001294 rpglen = pregion->vm_end - pregion->vm_start;
1295 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1296 rpgend = pregion->vm_pgoff + rpglen;
1297 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 continue;
1299
David Howells8feae132009-01-08 12:04:47 +00001300 /* handle inexactly overlapping matches between
1301 * mappings */
1302 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1303 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1304 /* new mapping is not a subset of the region */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001305 if (!(capabilities & NOMMU_MAP_DIRECT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 goto sharing_violation;
1307 continue;
1308 }
1309
David Howells8feae132009-01-08 12:04:47 +00001310 /* we've found a region we can share */
David Howells1e2ae592010-01-15 17:01:33 -08001311 pregion->vm_usage++;
David Howells8feae132009-01-08 12:04:47 +00001312 vma->vm_region = pregion;
1313 start = pregion->vm_start;
1314 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1315 vma->vm_start = start;
1316 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001318 if (pregion->vm_flags & VM_MAPPED_COPY)
David Howells8feae132009-01-08 12:04:47 +00001319 vma->vm_flags |= VM_MAPPED_COPY;
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001320 else {
David Howells8feae132009-01-08 12:04:47 +00001321 ret = do_mmap_shared_file(vma);
1322 if (ret < 0) {
1323 vma->vm_region = NULL;
1324 vma->vm_start = 0;
1325 vma->vm_end = 0;
David Howells1e2ae592010-01-15 17:01:33 -08001326 pregion->vm_usage--;
David Howells8feae132009-01-08 12:04:47 +00001327 pregion = NULL;
1328 goto error_just_free;
1329 }
1330 }
1331 fput(region->vm_file);
1332 kmem_cache_free(vm_region_jar, region);
1333 region = pregion;
1334 result = start;
1335 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 }
1337
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 /* obtain the address at which to make a shared mapping
1339 * - this is the hook for quasi-memory character devices to
1340 * tell us the location of a shared mapping
1341 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001342 if (capabilities & NOMMU_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 addr = file->f_op->get_unmapped_area(file, addr, len,
1344 pgoff, flags);
Namhyung Kimbb005a52011-05-24 17:11:27 -07001345 if (IS_ERR_VALUE(addr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 ret = addr;
Namhyung Kimbb005a52011-05-24 17:11:27 -07001347 if (ret != -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001348 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
1350 /* the driver refused to tell us where to site
1351 * the mapping so we'll have to attempt to copy
1352 * it */
Namhyung Kimbb005a52011-05-24 17:11:27 -07001353 ret = -ENODEV;
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001354 if (!(capabilities & NOMMU_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001355 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001357 capabilities &= ~NOMMU_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001358 } else {
1359 vma->vm_start = region->vm_start = addr;
1360 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 }
1362 }
1363 }
1364
David Howells8feae132009-01-08 12:04:47 +00001365 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
David Howells645d83c2009-09-24 15:13:10 +01001367 /* set up the mapping
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001368 * - the region is filled in if NOMMU_MAP_DIRECT is still set
David Howells645d83c2009-09-24 15:13:10 +01001369 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 if (file && vma->vm_flags & VM_SHARED)
David Howells8feae132009-01-08 12:04:47 +00001371 ret = do_mmap_shared_file(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 else
David Howells645d83c2009-09-24 15:13:10 +01001373 ret = do_mmap_private(vma, region, len, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 if (ret < 0)
David Howells645d83c2009-09-24 15:13:10 +01001375 goto error_just_free;
1376 add_nommu_region(region);
David Howells8feae132009-01-08 12:04:47 +00001377
Jie Zhangea637632009-12-14 18:00:02 -08001378 /* clear anonymous mappings that don't ask for uninitialized data */
1379 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1380 memset((void *)region->vm_start, 0,
1381 region->vm_end - region->vm_start);
1382
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001384 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 current->mm->total_vm += len >> PAGE_SHIFT;
1387
David Howells8feae132009-01-08 12:04:47 +00001388share:
1389 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
Mike Frysingercfe79c02010-01-06 17:23:23 +00001391 /* we flush the region from the icache only when the first executable
1392 * mapping of it is made */
1393 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1394 flush_icache_range(region->vm_start, region->vm_end);
1395 region->vm_icache_flushed = true;
1396 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397
Mike Frysingercfe79c02010-01-06 17:23:23 +00001398 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
David Howells8feae132009-01-08 12:04:47 +00001400 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
David Howells8feae132009-01-08 12:04:47 +00001402error_just_free:
1403 up_write(&nommu_region_sem);
1404error:
David Howells89a86402009-10-30 13:13:26 +00001405 if (region->vm_file)
1406 fput(region->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001407 kmem_cache_free(vm_region_jar, region);
David Howells89a86402009-10-30 13:13:26 +00001408 if (vma->vm_file)
1409 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001410 kmem_cache_free(vm_area_cachep, vma);
David Howells8feae132009-01-08 12:04:47 +00001411 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412
David Howells8feae132009-01-08 12:04:47 +00001413sharing_violation:
1414 up_write(&nommu_region_sem);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001415 pr_warn("Attempt to share mismatched mappings\n");
David Howells8feae132009-01-08 12:04:47 +00001416 ret = -EINVAL;
1417 goto error;
1418
1419error_getting_vma:
1420 kmem_cache_free(vm_region_jar, region);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001421 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1422 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001423 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 return -ENOMEM;
1425
David Howells8feae132009-01-08 12:04:47 +00001426error_getting_region:
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001427 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1428 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001429 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 return -ENOMEM;
1431}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001432
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001433SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1434 unsigned long, prot, unsigned long, flags,
1435 unsigned long, fd, unsigned long, pgoff)
1436{
1437 struct file *file = NULL;
1438 unsigned long retval = -EBADF;
1439
Al Viro120a7952010-10-30 02:54:44 -04001440 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001441 if (!(flags & MAP_ANONYMOUS)) {
1442 file = fget(fd);
1443 if (!file)
1444 goto out;
1445 }
1446
1447 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1448
Greg Ungererad1ed292012-06-04 14:29:59 +10001449 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001450
1451 if (file)
1452 fput(file);
1453out:
1454 return retval;
1455}
1456
Christoph Hellwiga4679372010-03-10 15:21:15 -08001457#ifdef __ARCH_WANT_SYS_OLD_MMAP
1458struct mmap_arg_struct {
1459 unsigned long addr;
1460 unsigned long len;
1461 unsigned long prot;
1462 unsigned long flags;
1463 unsigned long fd;
1464 unsigned long offset;
1465};
1466
1467SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1468{
1469 struct mmap_arg_struct a;
1470
1471 if (copy_from_user(&a, arg, sizeof(a)))
1472 return -EFAULT;
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001473 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001474 return -EINVAL;
1475
1476 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1477 a.offset >> PAGE_SHIFT);
1478}
1479#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1480
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481/*
David Howells8feae132009-01-08 12:04:47 +00001482 * split a vma into two pieces at address 'addr', a new vma is allocated either
1483 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 */
David Howells8feae132009-01-08 12:04:47 +00001485int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1486 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487{
David Howells8feae132009-01-08 12:04:47 +00001488 struct vm_area_struct *new;
1489 struct vm_region *region;
1490 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491
David Howells779c1022010-01-15 17:01:34 -08001492 /* we're only permitted to split anonymous regions (these should have
1493 * only a single usage on the region) */
1494 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +00001495 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
David Howells8feae132009-01-08 12:04:47 +00001497 if (mm->map_count >= sysctl_max_map_count)
1498 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
David Howells8feae132009-01-08 12:04:47 +00001500 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1501 if (!region)
1502 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503
David Howells8feae132009-01-08 12:04:47 +00001504 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1505 if (!new) {
1506 kmem_cache_free(vm_region_jar, region);
1507 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 }
David Howells8feae132009-01-08 12:04:47 +00001509
1510 /* most fields are the same, copy all, and then fixup */
1511 *new = *vma;
1512 *region = *vma->vm_region;
1513 new->vm_region = region;
1514
1515 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1516
1517 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001518 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001519 } else {
1520 region->vm_start = new->vm_start = addr;
1521 region->vm_pgoff = new->vm_pgoff += npages;
1522 }
1523
1524 if (new->vm_ops && new->vm_ops->open)
1525 new->vm_ops->open(new);
1526
1527 delete_vma_from_mm(vma);
1528 down_write(&nommu_region_sem);
1529 delete_nommu_region(vma->vm_region);
1530 if (new_below) {
1531 vma->vm_region->vm_start = vma->vm_start = addr;
1532 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1533 } else {
1534 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001535 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001536 }
1537 add_nommu_region(vma->vm_region);
1538 add_nommu_region(new->vm_region);
1539 up_write(&nommu_region_sem);
1540 add_vma_to_mm(mm, vma);
1541 add_vma_to_mm(mm, new);
1542 return 0;
1543}
1544
1545/*
1546 * shrink a VMA by removing the specified chunk from either the beginning or
1547 * the end
1548 */
1549static int shrink_vma(struct mm_struct *mm,
1550 struct vm_area_struct *vma,
1551 unsigned long from, unsigned long to)
1552{
1553 struct vm_region *region;
1554
David Howells8feae132009-01-08 12:04:47 +00001555 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1556 * and list */
1557 delete_vma_from_mm(vma);
1558 if (from > vma->vm_start)
1559 vma->vm_end = from;
1560 else
1561 vma->vm_start = to;
1562 add_vma_to_mm(mm, vma);
1563
1564 /* cut the backing region down to size */
1565 region = vma->vm_region;
David Howells1e2ae592010-01-15 17:01:33 -08001566 BUG_ON(region->vm_usage != 1);
David Howells8feae132009-01-08 12:04:47 +00001567
1568 down_write(&nommu_region_sem);
1569 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001570 if (from > region->vm_start) {
1571 to = region->vm_top;
1572 region->vm_top = region->vm_end = from;
1573 } else {
David Howells8feae132009-01-08 12:04:47 +00001574 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001575 }
David Howells8feae132009-01-08 12:04:47 +00001576 add_nommu_region(region);
1577 up_write(&nommu_region_sem);
1578
1579 free_page_series(from, to);
1580 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581}
1582
David Howells30340972006-09-27 01:50:20 -07001583/*
1584 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001585 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1586 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001587 */
David Howells8feae132009-01-08 12:04:47 +00001588int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589{
David Howells8feae132009-01-08 12:04:47 +00001590 struct vm_area_struct *vma;
Bob Liuf67d9b12011-05-24 17:12:56 -07001591 unsigned long end;
David Howells8feae132009-01-08 12:04:47 +00001592 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
Bob Liuf67d9b12011-05-24 17:12:56 -07001594 len = PAGE_ALIGN(len);
David Howells8feae132009-01-08 12:04:47 +00001595 if (len == 0)
1596 return -EINVAL;
1597
Bob Liuf67d9b12011-05-24 17:12:56 -07001598 end = start + len;
1599
David Howells8feae132009-01-08 12:04:47 +00001600 /* find the first potentially overlapping VMA */
1601 vma = find_vma(mm, start);
1602 if (!vma) {
Choi Gi-yongac714902014-04-07 15:37:36 -07001603 static int limit;
David Howells33e5d7692009-04-02 16:56:32 -07001604 if (limit < 5) {
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001605 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1606 current->pid, current->comm,
1607 start, start + len - 1);
David Howells33e5d7692009-04-02 16:56:32 -07001608 limit++;
1609 }
David Howells8feae132009-01-08 12:04:47 +00001610 return -EINVAL;
David Howells30340972006-09-27 01:50:20 -07001611 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612
David Howells8feae132009-01-08 12:04:47 +00001613 /* we're allowed to split an anonymous VMA but not a file-backed one */
1614 if (vma->vm_file) {
1615 do {
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001616 if (start > vma->vm_start)
David Howells8feae132009-01-08 12:04:47 +00001617 return -EINVAL;
David Howells8feae132009-01-08 12:04:47 +00001618 if (end == vma->vm_end)
1619 goto erase_whole_vma;
Namhyung Kimd75a3102011-05-24 17:11:25 -07001620 vma = vma->vm_next;
1621 } while (vma);
David Howells8feae132009-01-08 12:04:47 +00001622 return -EINVAL;
1623 } else {
1624 /* the chunk must be a subset of the VMA found */
1625 if (start == vma->vm_start && end == vma->vm_end)
1626 goto erase_whole_vma;
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001627 if (start < vma->vm_start || end > vma->vm_end)
David Howells8feae132009-01-08 12:04:47 +00001628 return -EINVAL;
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001629 if (offset_in_page(start))
David Howells8feae132009-01-08 12:04:47 +00001630 return -EINVAL;
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001631 if (end != vma->vm_end && offset_in_page(end))
David Howells8feae132009-01-08 12:04:47 +00001632 return -EINVAL;
David Howells8feae132009-01-08 12:04:47 +00001633 if (start != vma->vm_start && end != vma->vm_end) {
1634 ret = split_vma(mm, vma, start, 1);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001635 if (ret < 0)
David Howells8feae132009-01-08 12:04:47 +00001636 return ret;
David Howells8feae132009-01-08 12:04:47 +00001637 }
1638 return shrink_vma(mm, vma, start, end);
1639 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640
David Howells8feae132009-01-08 12:04:47 +00001641erase_whole_vma:
1642 delete_vma_from_mm(vma);
1643 delete_vma(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 return 0;
1645}
Paul Mundtb5073172007-07-21 04:37:25 -07001646EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647
Al Virobfce2812012-04-20 21:57:04 -04001648int vm_munmap(unsigned long addr, size_t len)
David Howells30340972006-09-27 01:50:20 -07001649{
Al Virobfce2812012-04-20 21:57:04 -04001650 struct mm_struct *mm = current->mm;
David Howells30340972006-09-27 01:50:20 -07001651 int ret;
David Howells30340972006-09-27 01:50:20 -07001652
1653 down_write(&mm->mmap_sem);
1654 ret = do_munmap(mm, addr, len);
1655 up_write(&mm->mmap_sem);
1656 return ret;
1657}
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001658EXPORT_SYMBOL(vm_munmap);
1659
1660SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1661{
Al Virobfce2812012-04-20 21:57:04 -04001662 return vm_munmap(addr, len);
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001663}
David Howells30340972006-09-27 01:50:20 -07001664
1665/*
David Howells8feae132009-01-08 12:04:47 +00001666 * release all the mappings made in a process's VM space
David Howells30340972006-09-27 01:50:20 -07001667 */
David Howells8feae132009-01-08 12:04:47 +00001668void exit_mmap(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669{
David Howells8feae132009-01-08 12:04:47 +00001670 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671
David Howells8feae132009-01-08 12:04:47 +00001672 if (!mm)
1673 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
David Howells8feae132009-01-08 12:04:47 +00001675 mm->total_vm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
David Howells8feae132009-01-08 12:04:47 +00001677 while ((vma = mm->mmap)) {
1678 mm->mmap = vma->vm_next;
1679 delete_vma_from_mm(vma);
1680 delete_vma(mm, vma);
Steven J. Magnani04c34962010-11-24 12:56:54 -08001681 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 }
1683}
1684
Linus Torvalds5d22fc22016-05-27 15:57:31 -07001685int vm_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686{
1687 return -ENOMEM;
1688}
1689
1690/*
David Howells6fa5f802006-09-27 01:50:21 -07001691 * expand (or shrink) an existing mapping, potentially moving it at the same
1692 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 *
David Howells6fa5f802006-09-27 01:50:21 -07001694 * under NOMMU conditions, we only permit changing a mapping's size, and only
David Howells8feae132009-01-08 12:04:47 +00001695 * as long as it stays within the region allocated by do_mmap_private() and the
1696 * block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 *
David Howells6fa5f802006-09-27 01:50:21 -07001698 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 */
Al Viro4b377ba2013-03-04 10:47:59 -05001700static unsigned long do_mremap(unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 unsigned long old_len, unsigned long new_len,
1702 unsigned long flags, unsigned long new_addr)
1703{
David Howells6fa5f802006-09-27 01:50:21 -07001704 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705
1706 /* insanity checks first */
Bob Liuf67d9b12011-05-24 17:12:56 -07001707 old_len = PAGE_ALIGN(old_len);
1708 new_len = PAGE_ALIGN(new_len);
David Howells8feae132009-01-08 12:04:47 +00001709 if (old_len == 0 || new_len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 return (unsigned long) -EINVAL;
1711
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001712 if (offset_in_page(addr))
David Howells8feae132009-01-08 12:04:47 +00001713 return -EINVAL;
1714
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 if (flags & MREMAP_FIXED && new_addr != addr)
1716 return (unsigned long) -EINVAL;
1717
David Howells8feae132009-01-08 12:04:47 +00001718 vma = find_vma_exact(current->mm, addr, old_len);
David Howells6fa5f802006-09-27 01:50:21 -07001719 if (!vma)
1720 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721
David Howells6fa5f802006-09-27 01:50:21 -07001722 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 return (unsigned long) -EFAULT;
1724
David Howells6fa5f802006-09-27 01:50:21 -07001725 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 return (unsigned long) -EPERM;
1727
David Howells8feae132009-01-08 12:04:47 +00001728 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 return (unsigned long) -ENOMEM;
1730
1731 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001732 vma->vm_end = vma->vm_start + new_len;
David Howells6fa5f802006-09-27 01:50:21 -07001733 return vma->vm_start;
1734}
1735
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001736SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1737 unsigned long, new_len, unsigned long, flags,
1738 unsigned long, new_addr)
David Howells6fa5f802006-09-27 01:50:21 -07001739{
1740 unsigned long ret;
1741
1742 down_write(&current->mm->mmap_sem);
1743 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1744 up_write(&current->mm->mmap_sem);
1745 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746}
1747
Michel Lespinasse240aade2013-02-22 16:35:56 -08001748struct page *follow_page_mask(struct vm_area_struct *vma,
1749 unsigned long address, unsigned int flags,
1750 unsigned int *page_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751{
Michel Lespinasse240aade2013-02-22 16:35:56 -08001752 *page_mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 return NULL;
1754}
1755
Bob Liu8f3b1322011-07-08 15:39:46 -07001756int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1757 unsigned long pfn, unsigned long size, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758{
Bob Liu8f3b1322011-07-08 15:39:46 -07001759 if (addr != (pfn << PAGE_SHIFT))
1760 return -EINVAL;
1761
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07001762 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001763 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764}
Luke Yang22c4af42006-07-14 00:24:09 -07001765EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766
Linus Torvalds3c0b9de2013-04-27 13:25:38 -07001767int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1768{
1769 unsigned long pfn = start >> PAGE_SHIFT;
1770 unsigned long vm_len = vma->vm_end - vma->vm_start;
1771
1772 pfn += vma->vm_pgoff;
1773 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1774}
1775EXPORT_SYMBOL(vm_iomap_memory);
1776
Paul Mundtf905bc42008-02-04 22:29:59 -08001777int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1778 unsigned long pgoff)
1779{
1780 unsigned int size = vma->vm_end - vma->vm_start;
1781
1782 if (!(vma->vm_flags & VM_USERMAP))
1783 return -EINVAL;
1784
1785 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1786 vma->vm_end = vma->vm_start + size;
1787
1788 return 0;
1789}
1790EXPORT_SYMBOL(remap_vmalloc_range);
1791
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1793 unsigned long len, unsigned long pgoff, unsigned long flags)
1794{
1795 return -ENOMEM;
1796}
1797
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798void unmap_mapping_range(struct address_space *mapping,
1799 loff_t const holebegin, loff_t const holelen,
1800 int even_cows)
1801{
1802}
Luke Yang22c4af42006-07-14 00:24:09 -07001803EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804
Nick Piggind0217ac2007-07-19 01:47:03 -07001805int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001806{
1807 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07001808 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08001809}
Paul Mundtb5073172007-07-21 04:37:25 -07001810EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07001811
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07001812void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1813{
1814 BUG();
1815}
1816EXPORT_SYMBOL(filemap_map_pages);
1817
Mike Frysingerf55f1992011-03-29 14:05:12 +01001818static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1819 unsigned long addr, void *buf, int len, int write)
David Howells0ec76a12006-09-27 01:50:15 -07001820{
David Howells0ec76a12006-09-27 01:50:15 -07001821 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07001822
1823 down_read(&mm->mmap_sem);
1824
1825 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07001826 vma = find_vma(mm, addr);
1827 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07001828 /* don't overrun this mapping */
1829 if (addr + len >= vma->vm_end)
1830 len = vma->vm_end - addr;
1831
1832 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07001833 if (write && vma->vm_flags & VM_MAYWRITE)
Jie Zhang79597222010-01-06 17:23:28 +00001834 copy_to_user_page(vma, NULL, addr,
1835 (void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07001836 else if (!write && vma->vm_flags & VM_MAYREAD)
Jie Zhang79597222010-01-06 17:23:28 +00001837 copy_from_user_page(vma, NULL, addr,
1838 buf, (void *) addr, len);
David Howells0ec76a12006-09-27 01:50:15 -07001839 else
1840 len = 0;
1841 } else {
1842 len = 0;
1843 }
1844
1845 up_read(&mm->mmap_sem);
Mike Frysingerf55f1992011-03-29 14:05:12 +01001846
1847 return len;
1848}
1849
1850/**
1851 * @access_remote_vm - access another process' address space
1852 * @mm: the mm_struct of the target address space
1853 * @addr: start address to access
1854 * @buf: source or destination buffer
1855 * @len: number of bytes to transfer
1856 * @write: whether the access is a write
1857 *
1858 * The caller must hold a reference on @mm.
1859 */
1860int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1861 void *buf, int len, int write)
1862{
1863 return __access_remote_vm(NULL, mm, addr, buf, len, write);
1864}
1865
1866/*
1867 * Access another process' address space.
1868 * - source/target buffer must be kernel space
1869 */
1870int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1871{
1872 struct mm_struct *mm;
1873
1874 if (addr + len < addr)
1875 return 0;
1876
1877 mm = get_task_mm(tsk);
1878 if (!mm)
1879 return 0;
1880
1881 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
1882
David Howells0ec76a12006-09-27 01:50:15 -07001883 mmput(mm);
1884 return len;
1885}
David Howells7e660872010-01-15 17:01:39 -08001886
1887/**
1888 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1889 * @inode: The inode to check
1890 * @size: The current filesize of the inode
1891 * @newsize: The proposed filesize of the inode
1892 *
1893 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1894 * make sure that that any outstanding VMAs aren't broken and then shrink the
1895 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1896 * automatically grant mappings that are too large.
1897 */
1898int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1899 size_t newsize)
1900{
1901 struct vm_area_struct *vma;
David Howells7e660872010-01-15 17:01:39 -08001902 struct vm_region *region;
1903 pgoff_t low, high;
1904 size_t r_size, r_top;
1905
1906 low = newsize >> PAGE_SHIFT;
1907 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1908
1909 down_write(&nommu_region_sem);
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001910 i_mmap_lock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08001911
1912 /* search for VMAs that fall within the dead zone */
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001913 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
David Howells7e660872010-01-15 17:01:39 -08001914 /* found one - only interested if it's shared out of the page
1915 * cache */
1916 if (vma->vm_flags & VM_SHARED) {
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001917 i_mmap_unlock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08001918 up_write(&nommu_region_sem);
1919 return -ETXTBSY; /* not quite true, but near enough */
1920 }
1921 }
1922
1923 /* reduce any regions that overlap the dead zone - if in existence,
1924 * these will be pointed to by VMAs that don't overlap the dead zone
1925 *
1926 * we don't check for any regions that start beyond the EOF as there
1927 * shouldn't be any
1928 */
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001929 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
David Howells7e660872010-01-15 17:01:39 -08001930 if (!(vma->vm_flags & VM_SHARED))
1931 continue;
1932
1933 region = vma->vm_region;
1934 r_size = region->vm_top - region->vm_start;
1935 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1936
1937 if (r_top > newsize) {
1938 region->vm_top -= r_top - newsize;
1939 if (region->vm_end > region->vm_top)
1940 region->vm_end = region->vm_top;
1941 }
1942 }
1943
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001944 i_mmap_unlock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08001945 up_write(&nommu_region_sem);
1946 return 0;
1947}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07001948
1949/*
1950 * Initialise sysctl_user_reserve_kbytes.
1951 *
1952 * This is intended to prevent a user from starting a single memory hogging
1953 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1954 * mode.
1955 *
1956 * The default value is min(3% of free memory, 128MB)
1957 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1958 */
1959static int __meminit init_user_reserve(void)
1960{
1961 unsigned long free_kbytes;
1962
1963 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1964
1965 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1966 return 0;
1967}
Paul Gortmakera4bc6fc2015-05-01 20:08:20 -04001968subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001969
1970/*
1971 * Initialise sysctl_admin_reserve_kbytes.
1972 *
1973 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1974 * to log in and kill a memory hogging process.
1975 *
1976 * Systems with more than 256MB will reserve 8MB, enough to recover
1977 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1978 * only reserve 3% of free pages by default.
1979 */
1980static int __meminit init_admin_reserve(void)
1981{
1982 unsigned long free_kbytes;
1983
1984 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1985
1986 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1987 return 0;
1988}
Paul Gortmakera4bc6fc2015-05-01 20:08:20 -04001989subsys_initcall(init_admin_reserve);